Re: Stacking context and offscreen bitmaps

I agree that the term 'offscreen bitmap' is an implementation detail but
there needs to be a term to signal how these stacking contexts are
different.

Maybe we can call them 'detached stacking context' to signal that they are
rendered separately from their ancestor. What do you think?

On Thu, Mar 7, 2013 at 1:21 PM, James Robinson <jamesr@google.com> wrote:

>
>
> On Thu, Mar 7, 2013 at 1:05 PM, Rik Cabanier <cabanier@gmail.com> wrote:
>
>> In order to have interoperable implementations of blending and filters,
>> there needs to be a specification that describes when offscreen bitmaps are
>> generated.
>
>
> No.  Specifications should be written in terms of observable behaviors,
> not in terms of implementation details.  It's fine to specify the desired
> rendering in such a way that using an offscreen bitmap is convenient to
> use, but whether or not an implementation uses offscreen bitmaps or not is
> not part of the spec.
>
>
>> Currently, the compositing spec states that every stacking context will
>> have a buffer [1] but that is not always the case [2]. This behavior is not
>> documented.
>>
>
>> I can see 2 ways forward:
>> 1. document the current browser behavior
>> Create a new document, or extend "Appendix E. Elaborate description of
>> Stacking Contexts" [3] so it describes when off-screen buffers should be
>> generated.
>> Browsers wouldn't have to change their behavior if they didn't follow the
>> spec except where it would make a difference.
>>
>
> That's why it is important to define the rendering you want to achieve,
> not the method in which that rendering is achieved.
>
>
>> For instance, if a browser did not create a buffer for a stacking context
>> but the spec said it should and there is blending or a filter that accesses
>> the backdrop, then the stacking context should create a buffer.
>> I think this would be fairly easy to implement but it would take quite a
>> bit of time to investigate and document. A drawback is that the spec would
>> be complex and hard to use for authors
>>
>> 2. Always give a stacking context on offscreen buffer if it contains an
>> element with blending.
>> This would be a bit harder to implement, but easy to specify.
>>
>
> What does "an element with blending" mean?  These things would be very
> expensive, whatever they are.
>
> - James
>
>
>> The biggest drawback is that it introduces more isolated groups. For
>> instance, 2d transform or 'position: abs' won't force isolation currently.
>>
>> Rik
>>
>>
>> 1:
>> https://dvcs.w3.org/hg/FXTF/rawfile/default/compositing/index.html#csscompositingrules_CSS
>> 2: http://lists.w3.org/Archives/Public/public-fx/2013JanMar/0109.html
>> 3: http://www.w3.org/TR/CSS21/zindex.html
>>
>
>